Deep Neural Networks for Time Series Momentum and Position Sizing

Summary

The paper presents Deep Momentum Networks, which combine deep learning with the volatility-scaling framework used in time series momentum. A neural network learns trend estimates and position sizes together, with its parameters trained to optimize the signal’s Sharpe ratio. This replaces the need to specify those components separately by hand.

Backtests across 88 continuous futures contracts report that an LSTM-based version achieved more than twice the Sharpe ratio of traditional methods before transaction costs, and continued to outperform with costs of up to 2–3 basis points. The authors also add turnover regularization so training can account for trading costs, including for less liquid assets. The evidence is limited to the reported backtests; the summary does not specify the evaluation period, benchmark details, or whether the results generalize beyond the tested contracts and cost assumptions.

Key ideas

• The network learns trend estimation and position sizing jointly within a volatility-scaled momentum framework.
• Training directly optimizes the Sharpe ratio of the resulting trading signal.
• Backtests on 88 continuous futures contracts report improved Sharpe ratios over traditional methods.
• A turnover regularization term allows the model to account for trading costs during training.

Tags

# Enhancing Time Series Momentum Strategies Using Deep Neural Networks


# Enhancing Time Series Momentum Strategies Using Deep Neural Networks









While time series momentum is a well-studied phenomenon in finance, common strategies require the explicit definition of both a trend estimator and a position sizing rule. In this paper, we introduce Deep Momentum Networks -- a hybrid approach which injects deep learning based trading rules into the volatility scaling framework of time series momentum. The model also simultaneously learns both trend estimation and position sizing in a data-driven manner, with networks directly trained by optimising the Sharpe ratio of the signal. Backtesting on a portfolio of 88 continuous futures contracts, we demonstrate that the Sharpe-optimised LSTM improved traditional methods by more than two times in the absence of transactions costs, and continue outperforming when considering transaction costs up to 2-3 basis points. To account for more illiquid assets, we also propose a turnover regularisation term which trains the network to factor in costs at run-time.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.